2790 papers
This paper proposes IR-PowerDet, a YOLOv12s-based framework incorporating multi-scale feature fusion, dynamic environment adaptation, and small-object enhancement modules to achieve robust, high-accuracy detection of infrared power equipment in complex and occluded scenarios, reaching a 96.8% mAP.
This paper proposes a statistically validated, density-aware hybrid framework combining XGBoost with unsupervised anomaly scoring to robustly detect malicious DNS-over-HTTPS tunneling, demonstrating significant improvements in recall and false negative reduction over baselines while confirming that deep generative components offer no additional benefit.
This paper presents an adaptive resource-aware MK-CKKS framework for privacy-preserving federated learning in IoT that achieves linear-complexity multi-key aggregation and significant performance gains through real-world experiments on MNIST and CIFAR-10, while transparently acknowledging that independent-key construction and physical hardware validation remain future work.
The paper introduces -Guard, a label-free drift detection method for visual recognition systems that achieves provable invariance to model recalibration by monitoring the rank agreement between predictive entropy and feature-space novelty scores, thereby avoiding the false positives that plague existing confidence-based monitors while maintaining sensitivity to genuine distribution shifts.
This paper identifies and quantifies "over-invariance" failures in geometric learning where excessive symmetry discards critical phase information, proposing Projective Character Pooling (PCP) as a targeted repair mechanism that effectively restores phase channels and improves diagnostic flatness compared to standard magnitude-only models, though it remains a scoped solution rather than a universal replacement for orbit averaging.
This paper proposes a measurement-driven adaptive data partitioning controller that dynamically reallocates sample budgets based on real-time training time and energy metrics to mitigate stragglers, reduce energy consumption, and improve training efficiency across heterogeneous distributed and federated learning systems.
FGD-Det is an efficient asymmetric multispectral object detection framework that addresses information density disparities and spatial misalignment through a heterogeneous dual-stream backbone and a stage-wise fusion strategy featuring Frequency-Guided Decoupled Alignment, achieving state-of-the-art performance on LLVIP and FLIR benchmarks with significantly reduced computational cost.
This paper proposes PR-CNN, a deep learning framework integrating convolutional neural networks, pyramid split attention, and a relation network to achieve highly accurate and robust recognition of bean leaf diseases by effectively addressing challenges such as subtle visual differences, complex backgrounds, and limited data availability.
This paper proposes AIO-MSVM-WMK, a novel all-in-one multi-class SVM framework that integrates weighted multiple kernel learning with sample optimization to achieve superior classification accuracy and significantly reduced computational time on high-dimensional datasets.
PredictStroke is a modular two-stage system that integrates an ensemble machine learning model for long-term stroke risk prediction with real-time, vision- and speech-based classifiers for acute symptom triage, demonstrating high performance in risk assessment while validating the feasibility of independent acute detection modules despite lower real-time scores.